Multi-Way Compression for Channel Neural Decoding with Quantization

The performance of model-driven channel neural decoding has surpassed that of traditional channel decoding algorithms, but at a higher complexity, making it difficult to implement on resource-constrained communication hardware. In this paper, we propose a quantization scheme for model-driven channel neural decoding, and combine the quantization scheme with TR decomposition and weight sharing algorithms to form different types of multi-way compression methods. Experimental results on LDPC, BCH and Hamming codes show that the proposed quantization and multi-way compression methods can effectively reduce the complexity of channel neural decoding without significant performance degradation.